MisoTTS on Apple Silicon: Local TTS Works, But Not in Real Time
A practical MisoTTS MLX run on local hardware: five voice samples, load time, generation speed, memory usage, audio outputs, and playback-ready blog samples.
Latest Articles
Browse all notes, tutorials, and experiments.
I Ran GPT Image Prompts on a Local Model. The First Results Were Embarrassing.
Same prompts, public GPT/Nano references, local HiDream-O1 generations: why the first images looked blurry, what fixed them, and where hosted models still win.
GPU Price History: Cloud vs Hardware
An interactive GPU pricing dashboard comparing NVIDIA generations, cloud rental rates, and hardware market prices from 2015 to 2026.
HiDream-I1 vs GPT-image prompts: a small benchmark note
I tried five public benchmark-style prompts on HiDream-I1, compared where it behaves like the public GPT image examples, and where it clearly still struggles.
Positional Encoding
Where Am I? This is the second part of our Transformer series blog, where we will deep dive into a topic called positional embedding. In this part, we’ll see what PE is and the types of PEs. Background Classical NL...
BazarAI Agent
As always, I am late to the party… I wanted to experiment with the agentic capacity of language models. The last time I used an agent was when I built the RAG from scratch. We used a search tool to build, but it was...
KV cache in transformers
“This will be a small experiment”. My thoughts as I started working on it a month ago. It is a simple concept of caching previous results for future calculations, similar to dynamic programming in DSA. I could not hav...
CI/CD with GitHub Actions and Google Cloud
In this era of AI copilots like GPT and Gemini, I decided — for reasons even I don’t fully understand — to build a CI/CD pipeline the old way: just me, Google, and a sea of Stack Overflow tabs. And somewhere between c...
Force Alignment: Montreal Force Aligner
In this comprehensive guide, I've outlined everything you need to know about force alignment using MFA. From understanding the basics to training your own custom models, this blog has you covered.
LoRA - Low Rank Adaptation
In the last part of Compression of NN[1], we delved into methods for reducing the memory footprint of large ML models, in this part let's look into how to fine-tune these large models in consumer grade resources with...
Unconditional image generation
In the realm of artistry, where creativity and imagination intertwine, a captivating scene often begins as a humble sketch on paper. Though fascinating in their monochrome simplicity, these lines and contours often le...
Unconditional image generation
I know I'm late for the party. It's been a long time since the diffusion models have been introduced. There have been massive improvements in image generation since I have known image generation with ML, from GANs to...
The road often traveled: tales of Chikkamagaluru and Rameswaram
Travel blog to Chikkamagaluru, Rameswaram. Madhuri, Netravati, Kuppalli, Dhanuskodi. #treak, #travelblog
From Vectors to Radiance: NeRF and Its Visual Wizardry
It reminds me of things I learned about vectors, lenses, calculus, or the properties of light, that I thought I'll never use.
Breaking Bits, a dive into bitwise NOT (~)
I was working on an ML model and I stumbled on the task of writing my loss function. Interestingly, I had to use a bitwise operator, among other operations, on an image. To convert it into a sketch. As I blindly plug...
Compression of deep neural networks: P-1
Since the inception of the Deep Neural Network, their complexity has been increasing. For a small comparison, look at the table below [1]. ArchitectureModel parameters (in Millions)Alex-Net60Basic-L2440 Alex-Net w...
Rajasthan, Where the dune meets the palace
A trip to Rajasthan, where you will explore from dunes to palaces. Udaipur lakes to Jaisalmer dunes and borders.
Rajasthan, where the dune meets the palace
A trip to Rajasthan, where you will explore from dunes to palaces. Udaipur lakes to Jaisalmer dunes and borders.
Rajasthan, where the dune meets the palace
A trip to Rajasthan, where you will explore from dunes to palaces. Udaipur lakes to Jaisalmer dunes and borders.
Rajasthan, where the dune meets the palace
A trip to Rajasthan, where you will explore from dunes to palaces. Udaipur lakes to Jaisalmer dunes and borders. Rajasthan, City palace, Food, Lake Pichola, Udaipur
Do floating-point numbers really float in your computer?
We started the blog with a meme then you know we are doing something out of the box. If you have ever worked with floating-point numbers. In any programming language, floating point arithmetic seems daunting. Especia...
First shine of the year 2023
#Honnavara #YEAR2023 On the new year's eve, our plan to celebrate 2023 had begun. Unsure of where to go or what to do, we scrambled all the information on the internet for a destination. We left from Hubli at around 7...
Does the language you speak influence the way you think?
If you are reading this, you probably know one or more language. Language is this awesome tool that facilitates exchange of ideas, thoughts. In fact, let me put a couple of bizarre English sentences before you and see...
Do floating-point numbers really float in your computer?
Learn how floating point, decimal , fraction are stored in the computer memory. With C, C++ program. Coading ,techinical writing.